EP0663632B1 - Method and apparatus for controlling a process - Google Patents

Method and apparatus for controlling a process Download PDF

Info

Publication number
EP0663632B1
EP0663632B1 EP94100589A EP94100589A EP0663632B1 EP 0663632 B1 EP0663632 B1 EP 0663632B1 EP 94100589 A EP94100589 A EP 94100589A EP 94100589 A EP94100589 A EP 94100589A EP 0663632 B1 EP0663632 B1 EP 0663632B1
Authority
EP
European Patent Office
Prior art keywords
model
layer thickness
coating system
air pressure
dependence
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Lifetime
Application number
EP94100589A
Other languages
German (de)
French (fr)
Other versions
EP0663632A1 (en
Inventor
Martin Dipl.-Ing. Niemann
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Siemens AG
Original Assignee
Siemens AG
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Siemens AG filed Critical Siemens AG
Priority to EP94100589A priority Critical patent/EP0663632B1/en
Priority to DE59404777T priority patent/DE59404777D1/en
Priority to AT94100589T priority patent/ATE161109T1/en
Priority to US08/373,819 priority patent/US5598329A/en
Priority to KR1019950000666A priority patent/KR100354410B1/en
Publication of EP0663632A1 publication Critical patent/EP0663632A1/en
Application granted granted Critical
Publication of EP0663632B1 publication Critical patent/EP0663632B1/en
Anticipated expiration legal-status Critical
Expired - Lifetime legal-status Critical Current

Links

Images

Classifications

    • CCHEMISTRY; METALLURGY
    • C23COATING METALLIC MATERIAL; COATING MATERIAL WITH METALLIC MATERIAL; CHEMICAL SURFACE TREATMENT; DIFFUSION TREATMENT OF METALLIC MATERIAL; COATING BY VACUUM EVAPORATION, BY SPUTTERING, BY ION IMPLANTATION OR BY CHEMICAL VAPOUR DEPOSITION, IN GENERAL; INHIBITING CORROSION OF METALLIC MATERIAL OR INCRUSTATION IN GENERAL
    • C23CCOATING METALLIC MATERIAL; COATING MATERIAL WITH METALLIC MATERIAL; SURFACE TREATMENT OF METALLIC MATERIAL BY DIFFUSION INTO THE SURFACE, BY CHEMICAL CONVERSION OR SUBSTITUTION; COATING BY VACUUM EVAPORATION, BY SPUTTERING, BY ION IMPLANTATION OR BY CHEMICAL VAPOUR DEPOSITION, IN GENERAL
    • C23C2/00Hot-dipping or immersion processes for applying the coating material in the molten state without affecting the shape; Apparatus therefor
    • C23C2/14Removing excess of molten coatings; Controlling or regulating the coating thickness
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F15/00Digital computers in general; Data processing equipment in general
    • G06F15/16Combinations of two or more digital computers each having at least an arithmetic unit, a program unit and a register, e.g. for a simultaneous processing of several programs
    • CCHEMISTRY; METALLURGY
    • C23COATING METALLIC MATERIAL; COATING MATERIAL WITH METALLIC MATERIAL; CHEMICAL SURFACE TREATMENT; DIFFUSION TREATMENT OF METALLIC MATERIAL; COATING BY VACUUM EVAPORATION, BY SPUTTERING, BY ION IMPLANTATION OR BY CHEMICAL VAPOUR DEPOSITION, IN GENERAL; INHIBITING CORROSION OF METALLIC MATERIAL OR INCRUSTATION IN GENERAL
    • C23CCOATING METALLIC MATERIAL; COATING MATERIAL WITH METALLIC MATERIAL; SURFACE TREATMENT OF METALLIC MATERIAL BY DIFFUSION INTO THE SURFACE, BY CHEMICAL CONVERSION OR SUBSTITUTION; COATING BY VACUUM EVAPORATION, BY SPUTTERING, BY ION IMPLANTATION OR BY CHEMICAL VAPOUR DEPOSITION, IN GENERAL
    • C23C2/00Hot-dipping or immersion processes for applying the coating material in the molten state without affecting the shape; Apparatus therefor
    • C23C2/14Removing excess of molten coatings; Controlling or regulating the coating thickness
    • C23C2/16Removing excess of molten coatings; Controlling or regulating the coating thickness using fluids under pressure, e.g. air knives
    • C23C2/18Removing excess of molten coatings from elongated material
    • C23C2/20Strips; Plates
    • CCHEMISTRY; METALLURGY
    • C23COATING METALLIC MATERIAL; COATING MATERIAL WITH METALLIC MATERIAL; CHEMICAL SURFACE TREATMENT; DIFFUSION TREATMENT OF METALLIC MATERIAL; COATING BY VACUUM EVAPORATION, BY SPUTTERING, BY ION IMPLANTATION OR BY CHEMICAL VAPOUR DEPOSITION, IN GENERAL; INHIBITING CORROSION OF METALLIC MATERIAL OR INCRUSTATION IN GENERAL
    • C23CCOATING METALLIC MATERIAL; COATING MATERIAL WITH METALLIC MATERIAL; SURFACE TREATMENT OF METALLIC MATERIAL BY DIFFUSION INTO THE SURFACE, BY CHEMICAL CONVERSION OR SUBSTITUTION; COATING BY VACUUM EVAPORATION, BY SPUTTERING, BY ION IMPLANTATION OR BY CHEMICAL VAPOUR DEPOSITION, IN GENERAL
    • C23C2/00Hot-dipping or immersion processes for applying the coating material in the molten state without affecting the shape; Apparatus therefor
    • C23C2/50Controlling or regulating the coating processes
    • C23C2/51Computer-controlled implementation
    • CCHEMISTRY; METALLURGY
    • C23COATING METALLIC MATERIAL; COATING MATERIAL WITH METALLIC MATERIAL; CHEMICAL SURFACE TREATMENT; DIFFUSION TREATMENT OF METALLIC MATERIAL; COATING BY VACUUM EVAPORATION, BY SPUTTERING, BY ION IMPLANTATION OR BY CHEMICAL VAPOUR DEPOSITION, IN GENERAL; INHIBITING CORROSION OF METALLIC MATERIAL OR INCRUSTATION IN GENERAL
    • C23CCOATING METALLIC MATERIAL; COATING MATERIAL WITH METALLIC MATERIAL; SURFACE TREATMENT OF METALLIC MATERIAL BY DIFFUSION INTO THE SURFACE, BY CHEMICAL CONVERSION OR SUBSTITUTION; COATING BY VACUUM EVAPORATION, BY SPUTTERING, BY ION IMPLANTATION OR BY CHEMICAL VAPOUR DEPOSITION, IN GENERAL
    • C23C2/00Hot-dipping or immersion processes for applying the coating material in the molten state without affecting the shape; Apparatus therefor
    • C23C2/50Controlling or regulating the coating processes
    • C23C2/52Controlling or regulating the coating processes with means for measuring or sensing
    • C23C2/525Speed of the substrate
    • GPHYSICS
    • G05CONTROLLING; REGULATING
    • G05BCONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/0265Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion
    • G05B13/027Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion using neural networks only
    • GPHYSICS
    • G05CONTROLLING; REGULATING
    • G05BCONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/04Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
    • G05B13/042Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
    • GPHYSICS
    • G05CONTROLLING; REGULATING
    • G05DSYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
    • G05D5/00Control of dimensions of material
    • G05D5/02Control of dimensions of material of thickness, e.g. of rolled material
    • G05D5/03Control of dimensions of material of thickness, e.g. of rolled material characterised by the use of electric means

Definitions

  • the invention relates to a method for guiding a coating system in which an output variable, which is dependent on a plurality of influencing variables acting on the coating system, is influenced by a manipulated variable acting on the coating system via an actuator such that it assumes the value of a predetermined control variable.
  • the invention further relates to a corresponding device.
  • EP 0 521 749 discloses a method for guiding a coating system, in which system, by means of a model of the coating system contained in a computing device, as a function of a target air pressure supplied to the model on the input side and influencing variables, on which the actual layer thickness is dependent and as a function of changeable model parameters a model output value is calculated , the model output value being compared with the measured actual value layer thickness and being changed as a function of the comparison result of the model parameters in the sense of reducing the deviation between the model output value and the actual layer thickness.
  • the model of the coating system adapted to the actual process is used to determine the target pressure as a function of the influencing variables.
  • this procedure does not always lead to the required high-quality coating.
  • the invention is therefore based on the object of optimally adapting the management of a coating system to the actual process.
  • the object is achieved by a method according to claim 1 or a device according to claim 11 and a method according to claim 2 or a device according to claim 12.
  • the process to be managed is thus represented by the model, which is adapted to the actual process by adaptation.
  • the manipulated variable is then calculated directly from the command variable, with a control device being no longer required at all if the process is sufficiently accurately represented by the model.
  • the invention thus offers the possibility of directly controlling a coating system with the reference variable without regulation, as a result of which the process is carried out more easily and quickly than is the case with regulation.
  • the reference variable for determining the manipulated variable (target air pressure) the reference variable (target layer thickness) is compared with the model output value and depending on the comparison result a value supplied to the model for the manipulated variable (target air pressure) is changed until the model output value of the reference variable (Target layer thickness) and that the value finally obtained for the manipulated variable (target air pressure) is fed to the actuator.
  • the manipulated variable is determined in accordance with claim 2 in such a way that the command variable (target layer thickness) and the influencing variables are fed to an inverse model of the coating system on the input side, which generates the manipulated variable (target air pressure) on the output side, also depending on the adaptation of the model of the coating system the inverse model is adapted to the process.
  • the inverse model of the coating system does not receive the manipulated variable as an input variable, but calculates the manipulated variable in the opposite direction from the other variables. Accordingly, the inverse model can be derived directly from the non-inverse model.
  • the model of the Coating system for determining the manipulated variable in this case the target air pressure, as a function of the actual layer thickness, ie the actual variable, is used.
  • the inverse model according to the article " Application state of artificial neural networks in automation technology” Difference between setpoint and actual value is supplied.
  • the inverse model according to the invention is supplied with the target layer thickness, that is to say the target size, and influencing variables. This results in a structure of the signal flow that differs from the control structure from Figure 5 in the article "Application status of artificial neural networks in automation technology".
  • control device for controlling the process is no longer necessary.
  • the control deviation between the model output value or the specified reference variable and the measured output variable of the process is additionally fed to a control device, the output signal of which is applied to the manipulated variable generated by means of the model of the process.
  • the computing device with the model of the process then acts as feedforward control for the control device, so that these only correct such control errors must not be covered by the model.
  • the control device is therefore only used to a small extent and can therefore be designed to be correspondingly simple.
  • the gain of the control device is advantageously set by means of the model output value supplied by the model of the process.
  • the model of the process is generated in a learning process, the model being learned first on the basis of existing knowledge about the process to be led.
  • the process is then controlled with this model or the corresponding inverse model.
  • the model Based on the measurement values of the manipulated variable, the influencing variables and the output variable of the process, the model can be improved and adapted to the real process.
  • the model output value is preferably formed by a - preferably additive - combination of a basic component and a correction component, the basic component being generated by a basic model of the process, which was created on the basis of existing knowledge of the process and based on the measured output variable of the process Process events are adapted, and wherein the correction component is generated by a correction device, which is adaptively adapted to the process events as a function of deviations between the model output value and the output quantity of the process which still exist after the basic model has been adapted.
  • the division of the model of the process into a basic model and a correction device results in the advantage that after a relatively short learning period, namely after the basic model has been adapted, a suitable means for controlling the process is available.
  • the process to be managed is thus well modeled by the basic model derived from the existing knowledge of the process and adapted to the current process, whereby, however, model errors remaining do not result from the attempt to adapt the model further, but are corrected by the correction device connected in parallel to the basic model.
  • the correction device connected in parallel to the basic model.
  • the basic model of the process is preferably implemented in the form of a neural network, in order to create the basic model of the process based on the existing knowledge of the process, for example in the form of characteristic curves which describe the relationship between the manipulated variable, the influencing variables and the output variable of the process and determined by measurements of these variables, a mathematical function is created that shows the dependency of the output variable of the process on the manipulated variable, the influencing variables and function parameters, and by realizing the function thus obtained as a neural network, the manipulated variable and the influencing variables on the input side are supplied and the network parameters corresponding to the functional parameters are adaptively adapted to the process.
  • the correction device is advantageously designed in such a way that that it provides reference point-related correction values for predetermined reference points in the multidimensional space defined by the manipulated variable and the influencing variables, that the reference point-related correction values are calculated in such a way from deviations between the model output value and the output variable of the process measured together with values of the manipulated variable and the influencing variables lying between the reference points that the respective measured deviation results again when the reference point-related correction values are interpolated, and that the correction component is calculated by interpolation between those reference point-related correction values whose reference points are adjacent to the values for the manipulated variable belonging to the correction component and the influencing variables.
  • correction values are initially only learned at a limited number of predetermined support points, so that the structure of the correction device can be correspondingly simple and the correction values can be learned quickly and easily at the few support points.
  • the correction component is then calculated for each arbitrary combination of values of the manipulated variable and the influencing variables in a simple manner by interpolation between the learned correction point-related correction values.
  • the interpolation for calculating the correction component is preferably carried out on the basis of fuzzy inferences by assigning a membership function to each support point, which has the value one at the relevant support point and up to the next neighboring support points Zero drops, and by weighting the reference point-related correction values with the associated assignment functions and superimposing them.
  • the correction component is preferably calculated in a neural network forming the correction device.
  • the method according to the invention is used in industrial process engineering processes which can be described by models based on existing knowledge of the respective process obtained, for example, by measuring process variables.
  • the method according to the invention regulates the zinc layer thickness of a strip passing through a bath with liquid zinc by blowing air from a nozzle against the strip emerging from the zinc bath, the zinc layer thickness being the starting variable of the process, the air pressure being the manipulated variable and the belt speed and the geometric arrangement of the nozzle with respect to the belt are the influencing variables influencing the process.
  • the device for guiding the coating system has corresponding means designed for carrying out the method according to the invention, i. H. in particular an appropriately programmed computing device.
  • FIG. 1 shows the diagram of a hot-dip galvanizing plant in which a rolled strip 1 passes through a bath 2 with liquid zinc at a speed v. After leaving the zinc bath 2, a still liquid zinc layer adheres to the belt 1. At a distance a from the belt 1 there is a nozzle 3, from which air flows against the belt 1 at a pressure p and thus removes excess zinc.
  • the air pressure p is set via an actuator 4, here a valve, as a function of an actuating variable p *.
  • a hot measuring device 5 is arranged behind the nozzle 3, with which the zinc layer thickness c on the strip 1 is measured.
  • Main influences for the zinc layer thickness c are the speed v of the belt 1, the air pressure p, the distance a between the belt 1 and the nozzle 3, the height h of the nozzle 3 above the zinc bath 2 and the angle ⁇ between the belt 1 and the nozzle 3rd
  • the aim is to set the manipulated variable p * so that the output variable c of the process, i.e. the actual measured zinc layer thickness, assumes the value of a reference variable c *, here the target zinc layer thickness.
  • the manipulated variable p * is carried out by an inverse model 7 of process 6, which receives the command variable c * and the influencing variables v, a, h and ⁇ as input variables and calculates the manipulated variable p * directly from these variables.
  • the inverse model 7 is derived directly from a model 8 of the process 6, which simulates the process 6 and accordingly, like the process, receives the manipulated variable p * and the influencing variables v, a, h and ⁇ as input variables.
  • these input variables are linked via modifiable model parameters m to form a model output value c M , which corresponds more closely to the real output variable c of process 6, the more precisely model 6 replicates process 6.
  • the model 8 pre-adapted on the basis of existing knowledge about the process 6 is adapted adaptively to the actual process after the entire device for guiding the process 6 has been put into operation.
  • the model output value c M is compared in a comparison device 9 with the measured output variable c of the process 6 and the deviation c M -c is fed to a learning algorithm represented by the circuit block 10, which accesses the model parameters m in the model 8 and these as a function of the determined deviation c M -c changed in the sense of a reduction of this deviation.
  • model parameters m are also changed in the inverse model 7, which only performs an inverse calculation in comparison to the model 8 and is therefore derived directly from the model 7 and also has the same model parameters m, so that the model 8 and the associated model inverse model 7 are both adapted to the actual process.
  • the manipulated variable p * is calculated by the inverse model 7 as a function of the reference variable c * and the influencing variables v, a, h and ⁇ so precisely that the output variable c of the process has the value of Takes c *.
  • the manipulated variable p * is calculated faster than with a conventional control, which also has to be constantly adapted to the actual process, is the case.
  • FIG. 3 shows an alternative exemplary embodiment for the device according to the invention for guiding process 6.
  • This exemplary embodiment differs from that in FIG. 2 by the lack of the inverse model 7.
  • control device 14 can additionally be provided, as shown in FIG. In FIG. 4, 15 denotes a computing device in which the circuit blocks 7 to 13 shown in FIGS. 2 and 3 are implemented.
  • the control deviation 14 is supplied on the input side with the control deviation ⁇ c formed in an additional comparison device 16 between the command variable c * or, as indicated by the broken line, the model output value c M and the output variable c of the process.
  • the output signal ⁇ p of the control device 14 is applied to the manipulated variable p * via an adder 17.
  • the computing device 15 with the model 8 thus forms a precontrol for the control device 14.
  • the control device 14 therefore only has to correct those control errors which are not covered by the model 8, so that the control device 14 is only stressed to a small extent and is accordingly simple can be trained.
  • the gain ⁇ c / ⁇ p of the control device 14 is set by means of the changes in the quantities c M and p * processed by the model 8.
  • model 8 The structure of the model 8 and the learning process for adapting the model 8 to the process are explained in more detail below. To simplify the illustration, only the influence of the belt speed v and the air pressure p on the zinc layer thickness c is considered.
  • model 8 shows the non-linear characteristic curves 18 for the zinc layer thickness c as a function of the strip speed v and the air pressure p. These characteristics should be modeled by model 8 of the process.
  • the model 8 consists of a basic model 19 and a correction device 20 parallel thereto, to which the input variables p and v are fed in each case.
  • the basic model 19 uses this to calculate a basic component c MF and the correction device 20 a correction component c MK , both of which are linked in an adder 21 to the model output value c M.
  • the variables v and p 3/2 are fed to the neural network 22 on the input side and are linked according to the mathematical function specified above by a summation node 23 and a node 24 with quotient function.
  • the connections of the nodes in the neural network 22 are each provided with a factor. If the connection is drawn thick, the factor can be learned (network parameters K1, K2, K3); otherwise the factor is constant (here 1.0).
  • the basic component c MF results at the output of the neural network 22.
  • the generated basic component c MF is compared in a comparison device 25 with the output variable c generated and measured by the process 6 as a function of the measured variables p and v.
  • the network parameters K1, K2 and K3 are changed in a learning algorithm 26 until the approximation of the basic component c MF to the output variable c of the process 6 cannot be further improved.
  • the basic component c MF generated by the basic model 19 thus adapted is shown in dashed lines in FIG. As can be seen, there are still differences between the output variable c of the process 6 and the basic component c MF provided by the basic model 19.
  • the correction component c MK serves to compensate for these differences, and its calculation is explained in more detail in the correction device 20 on the basis of FIG.
  • Correction values c MKij are now learned for these reference points (p i , v j ) by the difference ⁇ c M between the output variable c generated by the process 6 as a function of the measured variables p and v and the basic component c MF supplied by the basic model 19 is used to calculate the base point-related correction values c MKij .
  • the reference point-related correction values c MKij are calculated from the difference value ⁇ c M in such a way that, when the correction values c MKij are interpolated at the reference points (p i , v j ), the difference value ⁇ c M results for the current input variables p and v.
  • the correction component c MK is finally calculated by interpolation between the correction point values c MKij learned in this way.
  • the interpolation takes place on the basis of fuzzy inferences, for which each reference point (p i , v j ) is assigned an assignment or membership function 27, which weights the learned correction value c MKij at the relevant reference point and in its vicinity w weights.
  • the assignment functions 27 have the value 1 at the associated reference points (p i , v j ), which drops linearly to the value 0 up to the respectively adjacent reference points.
  • FIG. 9 shows an example of the implementation of the above calculation rule for the correction component c MK through a neural network 28.
  • This receives as inputs the variables v and p, which are fed to a first layer with nodes 29 with sigmoid functions.
  • the triangular membership functions 27 shown in FIG. 8 are each formed by two of the nodes 29 and subsequent addition of the node output signals in a summation node 30 each.
  • the values w 1 to w 4 which denote the quantities p and v weighted with the membership functions 27, therefore appear at the outputs of the summation nodes 30.

Landscapes

  • Chemical & Material Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Organic Chemistry (AREA)
  • Metallurgy (AREA)
  • Mechanical Engineering (AREA)
  • Materials Engineering (AREA)
  • Chemical Kinetics & Catalysis (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Evolutionary Computation (AREA)
  • Computer Hardware Design (AREA)
  • Software Systems (AREA)
  • Medical Informatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • General Engineering & Computer Science (AREA)
  • Feedback Control In General (AREA)
  • Electrical Discharge Machining, Electrochemical Machining, And Combined Machining (AREA)
  • Threshing Machine Elements (AREA)

Abstract

The method concerns control of a process (6) whose output (c) - which depends on several quantities (v,a,h) influencing the process-is adjusted by means of a control variable (p<*>) in such a way that it becomes equal to a predetermined input (c<*>). A computer process model (8)- whose input include the variable (p<*>), the quantities (v,a,h,) and variable model parameters (m)-is used to calculate a model output (cm) which is compared with the measured output (c) of the process. Dependent on the comparison result, the model parameters are varied to minimise the discrepancy between (cm) and (c), and to use the model adapted to the actual process conditions for adjustment of the control variable (p<*>). The appts. include a computer provided with a learning algorithm for variation of the model parameters, and means - which include an inverse model of the process - for determining (p<*>).

Description

Die Erfindung betrifft ein Verfahren zur Führung einer Beschichtungsanlage, bei dem eine Ausgangsgröße, die von mehreren, auf die Beschichtungsanlage einwirkenden Einflußgrößen abhängig ist, durch eine über ein Stellglied auf die Beschichtungsanlage einwirkende Stellgröße derart beeinflußt wird, daß sie den Wert einer vorgegebenen Führungsgröße annimmt. Die Erfindung betrifft ferner eine entsprechende Vorrichtung.The invention relates to a method for guiding a coating system in which an output variable, which is dependent on a plurality of influencing variables acting on the coating system, is influenced by a manipulated variable acting on the coating system via an actuator such that it assumes the value of a predetermined control variable. The invention further relates to a corresponding device.

Es ist allgemein bekannt, einen solchen Prozeß derart zu führen, daß die Ausgangsgröße des Prozesses, also die Regelgröße, mit einer Führungsgröße verglichen wird und die so erhaltene Regelabweichung einer Regeleinrichtung zugeführt wird, die ausgangsseitig eine Stellgröße erzeugt, mit der der Prozeß über ein Stellglied beeinflußt wird. Neben der Stellgröße wirken auch eine Reihe anderer Einflußgrößen auf den Prozeß, wobei der Zusammenhang zwischen der Stellgröße und den Einflußgrößen einerseits und der Ausgangsgröße des Prozesses andererseits durch Kennlinien dargestellt werden kann, welche durch Messungen der Größen erhalten werden. Diese in der Praxis fast immer nichtlinearen Kennlinien sind jedoch meist nicht genau bekannt und können sich aufgrund von sich ändernden Prozeßbedingungen verändern. Dabei ergibt sich das Problem der Anpassung der Regelung des Prozesses an die sich ändernden Prozeßbedingungen, um zu erreichen, daß die Regeleinrichtung immer im optimalen Arbeitspunkt arbeitet. Aus der EP 0 521 749 ist ein Verfahren zur Führung einer Beschichtungsanlage offenbart, wobei in dieser Anlage mittels eines in einer Recheneinrichtung enthaltenen Modells der Beschichtungsanlage in Abhängigkeit von einem dem Modell eingangsseitig zugeführten Solluftdruck und Einflußgrößen, von denen die Istschichtdicke abhängig ist sowie in Abhängigkeit von veränderbaren Modellparametern ein Modellausgangswert berechnet wird, wobei der Modellausgangswert mit der gemessenen Istwertschichtdicke verglichen und in Abhängigkeit von dem Vergleichsergebnis der Modellparameter im Sinne einer Verringerung der Abweichung zwischen Modellausgangswert und der Istschichtdicke verändert wird. Dabei wird das so an das tatsächliche Prozeßgeschehen adaptierte Modell der Beschichtungsanlage zur Bestimmung des Solldrucks in Abhängigkeit von den Einflußgrößen herangezogen. Diese Vorgehensweise führt jedoch nicht immer zu der geforderten qualitativ hochwertigen Beschichtung.It is generally known to conduct such a process in such a way that the output variable of the process, that is to say the controlled variable, is compared with a reference variable and the control deviation obtained in this way is fed to a control device which generates an actuating variable on the output side with which the process operates via an actuator being affected. In addition to the manipulated variable, a number of other influencing variables also act on the process, the relationship between the manipulated variable and the influencing variables on the one hand and the output variable of the process on the other hand being able to be represented by characteristic curves which are obtained by measurements of the variables. However, these characteristic curves, which are almost always non-linear in practice, are usually not exactly known and can change due to changing process conditions. The problem arises of adapting the control of the process to the changing process conditions in order to ensure that the control device always works at the optimum operating point. EP 0 521 749 discloses a method for guiding a coating system, in which system, by means of a model of the coating system contained in a computing device, as a function of a target air pressure supplied to the model on the input side and influencing variables, on which the actual layer thickness is dependent and as a function of changeable model parameters a model output value is calculated , the model output value being compared with the measured actual value layer thickness and being changed as a function of the comparison result of the model parameters in the sense of reducing the deviation between the model output value and the actual layer thickness. The model of the coating system adapted to the actual process is used to determine the target pressure as a function of the influencing variables. However, this procedure does not always lead to the required high-quality coating.

Der Erfindung liegt daher die Aufgabe zugrunde, die Führung einer Beschichtungsanlage in optimaler Weise an das tatsächliche Prozeßgeschehen anzupassen.The invention is therefore based on the object of optimally adapting the management of a coating system to the actual process.

Gemäß der Erfindung wird die Aufgabe durch ein Verfahren gemäß Anspruch 1 bzw. eine Vorrichtung gemäß Anspruch 11 sowie ein Verfahren gemäß Anspruch 2 bzw. eine Vorrichtung gemäß Anspruch 12 gelöst.According to the invention, the object is achieved by a method according to claim 1 or a device according to claim 11 and a method according to claim 2 or a device according to claim 12.

Der zu führende Prozeß wird also durch das Modell abgebildet, wobei dieses durch Adaption an das tatsächliche Prozeßgeschehen angepaßt wird. Mit dem adaptierten Modell wird dann aus der Führungsgröße die Stellgröße direkt berechnet, wobei bei hinreichend genauer Abbildung des Prozesses durch das Modell eine Regeleinrichtung überhaupt nicht mehr erforderlich ist. Die Erfindung bietet also die Möglichkeit, eine Beschichtungsanlage mit der Führungsgröße ohne Regelung direkt zu steuern, wodurch die Prozeßführung einfacher und schneller erfolgt, als dies bei einer Regelung der Fall ist.The process to be managed is thus represented by the model, which is adapted to the actual process by adaptation. With the adapted model, the manipulated variable is then calculated directly from the command variable, with a control device being no longer required at all if the process is sufficiently accurately represented by the model. The invention thus offers the possibility of directly controlling a coating system with the reference variable without regulation, as a result of which the process is carried out more easily and quickly than is the case with regulation.

Entsprechend Anspruch 1 ist vorgesehen, daß zur Bestimmung der Stellgröße (Solluftdruck) die Führungsgröße (Sollschichtdicke) mit dem Modellausgangswert verglichen wird und in Abhängigkeit von dem Vergleichsergebnis ein dem Modell zugeführter Wert für die Stellgröße (Solluftdruck) solange verändert wird, bis der Modellausgangswert der Führungsgröße (Sollschichtdicke) entspricht, und daß der schließlich erhaltene Wert für die Stellgröße (Solluftdruck) dem Stellglied zugeführt wird.According to claim 1 it is provided that for determining the manipulated variable (target air pressure) the reference variable (target layer thickness) is compared with the model output value and depending on the comparison result a value supplied to the model for the manipulated variable (target air pressure) is changed until the model output value of the reference variable (Target layer thickness) and that the value finally obtained for the manipulated variable (target air pressure) is fed to the actuator.

Die Bestimmung der Stellgröße erfolgt gemäß Anspruch 2 in der Weise, daß die Führungsgröße (Sollschichtdicke) und die Einflußgrößen eingangsseitig einem inversen Modell der Beschichtungsanlage zugeführt werden, das ausgangsseitig die Stellgröße (Solluftdruck) erzeugt, wobei in Abhängigkeit von der Adaption des Modells der Beschichtungsanlage auch das inverse Modell an das Prozeßgeschehen angepaßt wird. Das inverse Modell der Beschichtungsanlage erhält im Unterschied zu dem nichtinversen Modell die Stellgröße nicht als Eingangsgröße, sondern berechnet die Stellgröße in umgekehrter Richtung aus den anderen Größen. Dementsprechend läßt sich das inverse Modell direkt aus dem nichtinversen Modell ableiten. Zwar ist der grundsätzliche Einsatz inverser Modelle aus dem Artikel ,,Anwendungsstand Künstlicher Neuronaler Netze in der Automatisierungstechnik" von Neumerkel und Lohnert, ATP Automatisierungstechnische Praxis 34 (1992) November, No. 11 bekannt, doch erfolgt dieser Einsatz in grundsätzlich unterschiedlicher Weise zu dem Einsatz des inversen Modells gemäß der Erfindung. Während das in dem Artikel ,,Anwendungsstand Künstlicher Neuronaler Netze in der Automatisierungstechnik" das inverse Modell ein festes Modell ist, wird das inverse Modell gemäß dem erfindungsgemäßen Einsatz an das Prozeßgeschehen adaptiert. Außerdem erfolgt das Trainieren des inversen Modells nicht wie in dem Artikel ,,Anwendungsstand Künstlicher Neuronaler Netze in der Automatisierungstechnik", sondern in Abhängigkeit von der Adaption des nichtinversen Modells der Beschichtungsanlage. Ein weiterer Unterschied besteht darin, daß das an das tatsächliche Prozeßgeschehen adaptierte Modell der Beschichtungsanlage zur Bestimmung der Stellgröße, in diesem Fall des Solluftdrucks, in Abhängigkeit von der Istschichtdicke, d.h. der Istgröße, herangezogen wird. Ein weiterer Unterschied besteht darin, daß dem inversen Modell gemäß dem Artikel "Anwendungsstand Künstlicher Neuronaler Netze in der Automatisierungstechnik" die Differenz zwischen Soll- und Istwert zugeführt wird. Demgegenüber werden dem inversen Modell gemäß der Erfindung die Sollschichtdicke, also die Sollgröße, sowie Einflußgrößen zugeführt. Auf diese Weise ergibt sich eine von der Regelungsstruktur aus Bild 5 im Artikel ,,Anwendungsstand Künstlicher Neuronaler Netze in der Automatisierungstechnik" verschiedene Struktur des Signalflusses.The manipulated variable is determined in accordance with claim 2 in such a way that the command variable (target layer thickness) and the influencing variables are fed to an inverse model of the coating system on the input side, which generates the manipulated variable (target air pressure) on the output side, also depending on the adaptation of the model of the coating system the inverse model is adapted to the process. In contrast to the non-inverse model, the inverse model of the coating system does not receive the manipulated variable as an input variable, but calculates the manipulated variable in the opposite direction from the other variables. Accordingly, the inverse model can be derived directly from the non-inverse model. Although the basic use of inverse models is known from the article "State of application of artificial neural networks in automation technology" by Neumerkel and Lohnert, ATP automation technology practice 34 (1992) November, No. 11, this use is fundamentally different from the use of the inverse model according to the invention. While the inverse model in the article "State of application of artificial neural networks in automation technology" is a fixed model, the inverse model is adapted to the process in accordance with the use according to the invention. In addition, the training of the inverse model does not take place as in the article "Application status of artificial neural networks in automation technology", but in dependence on the adaptation of the non-inverse model of the coating system. Another difference is that the model of the Coating system for determining the manipulated variable, in this case the target air pressure, as a function of the actual layer thickness, ie the actual variable, is used.Another difference is that the inverse model according to the article " Application state of artificial neural networks in automation technology" Difference between setpoint and actual value is supplied. In contrast, the inverse model according to the invention is supplied with the target layer thickness, that is to say the target size, and influencing variables. This results in a structure of the signal flow that differs from the control structure from Figure 5 in the article "Application status of artificial neural networks in automation technology".

Wie bereits erwähnt wurde, ist bei hinreichend genauer Nachbildung des zu führenden Prozesses durch das Modell eine Regeleinrichtung zur Regelung des Prozesses nicht mehr notwendig. Für den Fall, daß dennoch eine Regelung als erforderlich angesehen wird, wird zusätzlich die Regelabweichung zwischen dem Modellausgangswert oder der vorgegebenen Führungsgröße und der gemessenen Ausgangsgröße des Prozesses einer Regeleinrichtung zugeführt, deren Ausgangssignal der mittels des Modells des Prozesses erzeugten Stellgröße aufgeschaltet wird. Die Recheneinrichtung mit dem Modell des Prozesses wirkt dann als Vorsteuerung für die Regeleinrichtung, so daß diese nur noch solche Regelfehler ausregeln muß, die von dem Modell nicht erfaßt werden. Die Regeleinrichtung wird daher nur in geringem Maße beansprucht und kann somit entsprechend einfach ausgebildet sein.As already mentioned, if the model to be replicated with sufficient accuracy, a control device for controlling the process is no longer necessary. In the event that regulation is still considered necessary, the control deviation between the model output value or the specified reference variable and the measured output variable of the process is additionally fed to a control device, the output signal of which is applied to the manipulated variable generated by means of the model of the process. The computing device with the model of the process then acts as feedforward control for the control device, so that these only correct such control errors must not be covered by the model. The control device is therefore only used to a small extent and can therefore be designed to be correspondingly simple.

Um zu erreichen, daß die Regeleinrichtung immer im optimalen Arbeitspunkt arbeitet, wird in vorteilhafter Weise die Verstärkung der Regeleinrichtung mittels des von dem Modell des Prozesses gelieferten Modellausgangswertes eingestellt.In order to ensure that the control device always works at the optimum operating point, the gain of the control device is advantageously set by means of the model output value supplied by the model of the process.

Das Modell des Prozesses wird in einem Lernprozeß generiert, wobei zunächst aufgrund von vorhandenem Wissen über den zu führenden Prozeß das Modell gelernt wird. Mit diesem Modell bzw. dem entsprechenden inversen Modell wird dann der Prozeß gesteuert. Anhand der nun anfallenden Meßwerte der Stellgröße, der Einflußgrößen und der Ausgangsgröße des Prozesses läßt sich das Modell verbessern und an das reale Prozeßgeschehen anpassen. Dabei wird vorzugsweise der Modellausgangswert durch eine - vorzugsweise additive - Verknüpfung einer Grundkomponente und einer Korrekturkomponente gebildet, wobei die Grundkomponente von einem Grundmodell des Prozesses erzeugt wird, das aufgrund von vorhandenem Wissen über den Prozeß erstellt worden ist und anhand der gemessenen Ausgangsgröße des Prozesses an das Prozeßgeschehen adaptiert wird, und wobei die Korrekturkomponente von einer Korrektureinrichtung erzeugt wird, die in Abhängigkeit von nach erfolgter Adaption des Grundmodells noch bestehenden Abweichungen zwischen dem Modellausgangswert und der Ausgangsgröße des Prozesses adaptiv an das Prozeßgeschehen angepaßt wird. Aus der Aufteilung des Modells des Prozesses in ein Grundmodell und eine Korrektureinrichtung ergibt sich der Vorteil, daß bereits nach relativ kurzer Lernzeit, nämlich nach erfolgter Adaption des Grundmodells, ein geeignetes Mittel zur Steuerung des Prozesses zur Verfügung steht. Der zu führende Prozeß wird also durch das aus dem vorhandenen Wissen über den Prozeß abgeleitete und an das aktuelle Prozeßgeschehen angepaßte Grundmodell gut nachgebildet, wobei jedoch darüber hinaus verbleibende Modellfehler nicht durch den Versuch einer weiteren adaptiven Anpassung des Modells, sondern durch die parallel zu dem Grundmodell geschaltete Korrektureinrichtung korrigiert werden. Gegenüber einer ausschließlichen Verwendung eines Modells zur Erzeugung des Modellausgangswerts ergibt sich der Vorteil, daß sowohl der Aufbau als auch das Training des Grundmodells und der Korrektureinrichtung einfacher ist, als bei einem einzigen Modell, das dieselben Anforderungen erfüllen soll.The model of the process is generated in a learning process, the model being learned first on the basis of existing knowledge about the process to be led. The process is then controlled with this model or the corresponding inverse model. Based on the measurement values of the manipulated variable, the influencing variables and the output variable of the process, the model can be improved and adapted to the real process. The model output value is preferably formed by a - preferably additive - combination of a basic component and a correction component, the basic component being generated by a basic model of the process, which was created on the basis of existing knowledge of the process and based on the measured output variable of the process Process events are adapted, and wherein the correction component is generated by a correction device, which is adaptively adapted to the process events as a function of deviations between the model output value and the output quantity of the process which still exist after the basic model has been adapted. The division of the model of the process into a basic model and a correction device results in the advantage that after a relatively short learning period, namely after the basic model has been adapted, a suitable means for controlling the process is available. The process to be managed is thus well modeled by the basic model derived from the existing knowledge of the process and adapted to the current process, whereby, however, model errors remaining do not result from the attempt to adapt the model further, but are corrected by the correction device connected in parallel to the basic model. Compared to the exclusive use of a model for generating the model starting value, there is the advantage that both the construction and the training of the basic model and the correction device are simpler than in the case of a single model which is intended to meet the same requirements.

Das Grundmodell des Prozesses wird vorzugsweise in Form eines neuronalen Netzwerkes realisiert, indem zur Erstellung des Grundmodells des Prozesses aufgrund des vorhandenen Wissens über den Prozeß, beispielsweise in Form von Kennlinien, die den Zusammenhang zwischen der Stellgröße, den Einflußgrößen und der Ausgangsgröße des Prozesses beschreiben und durch Messungen dieser Größen ermittelt werden, eine mathematische Funktion erstellt wird, die die Abhängigkeit der Ausgangsgröße des Prozesses von der Stellgröße, den Einflußgrößen und von Funktionsparametern wiedergibt, und indem die so erhaltene Funktion als neuronales Netzwerk realisiert wird, dem eingangsseitig die Stellgröße und die Einflußgrößen zugeführt werden und dessen den Funktionsparametern entsprechende Netzwerkparameter adaptiv an das Prozeßgeschehen angepaßt werden. Hierdurch wird erreicht, daß von vorneherein die Struktur des neuronalen Netzwerkes anhand des vorhandenen Wissens über den zu führenden Prozeß, nämlich die gemessenen Kennlinien und die daraus abgeleitete mathematische Funktion, an den Prozeß angepaßt ist, so daß das neuronale Netzwerk im Vergleich zu einem allgemeinen Netzwerk einfach aufgebaut und über seine Netzwerkparameter entsprechend einfach und schnell an das reale Prozeßgeschehen adaptiert werden kann.The basic model of the process is preferably implemented in the form of a neural network, in order to create the basic model of the process based on the existing knowledge of the process, for example in the form of characteristic curves which describe the relationship between the manipulated variable, the influencing variables and the output variable of the process and determined by measurements of these variables, a mathematical function is created that shows the dependency of the output variable of the process on the manipulated variable, the influencing variables and function parameters, and by realizing the function thus obtained as a neural network, the manipulated variable and the influencing variables on the input side are supplied and the network parameters corresponding to the functional parameters are adaptively adapted to the process. This ensures that the structure of the neural network is adapted to the process from the outset based on the available knowledge of the process to be guided, namely the measured characteristic curves and the mathematical function derived therefrom, so that the neural network is compared to a general network simply constructed and can be easily and quickly adapted to the real process via its network parameters.

Verbleibende Modell fehler, die daraus resultieren, daß das Grundmodell bzw. das neuronale Netzwerk wegen seines einfachen Aufbaus nur bis zu einer bestimmten Grenze an das Prozeßgeschehen angepaßt werden kann, werden, wie bereits erläutert, durch die Korrektureinrichtung korrigiert. Die Korrektureinrichtung ist in vorteilhafter Weise derart ausgebildet, daß sie für vorgegebene Stützstellen in dem von der Stellgröße und den Einflußgrößen definierten mehrdimensionalen Raum stützstellenbezogene Korrekturwerte bereitstellt, daß die stützstellenbezogenen Korrekturwerte derart aus Abweichungen zwischen dem Modellausgangswert und der zusammen mit zwischen den Stützpunkten liegenden Werten der Stellgröße und der Einflußgrößen gemessenen Ausgangsgröße des Prozesses berechnet wird, daß sich bei Interpolation der stützstellenbezogenen Korrekturwerte die jeweilige gemessene Abweichung wieder ergibt, und daß die Korrekturkomponente durch Interpolation zwischen denjenigen stützstellenbezogenen Korrekturwerten berechnet wird, deren Stützstellen den Werten für die zu der Korrekturkomponente gehörende Stellgröße und die Einflußgrößen benachbart sind. Anhand von gemessenen Modellfehlern werden also zunächst Korrekturwerte nur an einer begrenzten Anzahl von vorgegebenen Stützstellen gelernt, so daß der Aufbau der Korrektureinrichtung entsprechend einfach sein kann und das Lernen der Korrekturwerte an den wenigen Stützstellen einfach und schnell erfolgt. Die Korrekturkomponente wird dann für jede beliebig auftretende Wertekombination der Stellgröße und der Einflußgrößen in einfacher Weise durch Interpolation zwischen den gelernten stützstellenbezogenen Korrekturwerten berechnet.Remaining model errors which result from the fact that the basic model or the neural network can only be adapted to the process process up to a certain limit because of its simple structure are, as already explained, corrected by the correction device. The correction device is advantageously designed in such a way that that it provides reference point-related correction values for predetermined reference points in the multidimensional space defined by the manipulated variable and the influencing variables, that the reference point-related correction values are calculated in such a way from deviations between the model output value and the output variable of the process measured together with values of the manipulated variable and the influencing variables lying between the reference points that the respective measured deviation results again when the reference point-related correction values are interpolated, and that the correction component is calculated by interpolation between those reference point-related correction values whose reference points are adjacent to the values for the manipulated variable belonging to the correction component and the influencing variables. On the basis of measured model errors, correction values are initially only learned at a limited number of predetermined support points, so that the structure of the correction device can be correspondingly simple and the correction values can be learned quickly and easily at the few support points. The correction component is then calculated for each arbitrary combination of values of the manipulated variable and the influencing variables in a simple manner by interpolation between the learned correction point-related correction values.

Dabei erfolgt die Interpolation zur Berechnung der Korrekturkomponente vorzugsweise auf der Grundlage von Fuzzy-Folgerungen, indem jeder Stützstelle eine Zuordnungs- (Membership-)Funktion zugeordnet wird, die an der betreffenden Stützstelle den Wert Eins aufweist und bis zu den nächsten benachbarten Stützstellen auf den Wert Null abfällt, und indem die stützstellenbezogenen Korrekturwerte mit den zugehörigen Zuordnungsfunktionen gewichtet und einander überlagert werden.The interpolation for calculating the correction component is preferably carried out on the basis of fuzzy inferences by assigning a membership function to each support point, which has the value one at the relevant support point and up to the next neighboring support points Zero drops, and by weighting the reference point-related correction values with the associated assignment functions and superimposing them.

Wie auch schon die Grundkomponente, wird auch die Korrekturkomponente bevorzugt in einem die Korrektureinrichtung bildenden neuronalen Netzwerk berechnet.Like the basic component, the correction component is preferably calculated in a neural network forming the correction device.

Das erfindungsgemäße Verfahren findet bei industriellen verfahrenstechnischen Prozessen Anwendung, die aufgrund von vorhandenem, beispielsweise durch Messung von Prozeßgrößen erhaltenen Wissen über den jeweiligen Prozeß durch Modelle beschrieben werden können. In diesem Zusammenhang wird mit dem erfindungsgemäßen Verfahren die Zinkschichtdicke eines ein Bad mit flüssigem Zink durchlaufenden Bandes geregelt, indem aus einer Düse Luft gegen das aus dem Zinkbad austretende Band geblasen wird, wobei die Zinkschichtdicke die Ausgangsgröße des Prozesses ist, der Luftdruck die Stellgröße bildet und die Bandgeschwindigkeit sowie die geometrische Anordnung der Düse in bezug auf das Band die auf den Prozeß einwirkenden Einflußgrößen sind.The method according to the invention is used in industrial process engineering processes which can be described by models based on existing knowledge of the respective process obtained, for example, by measuring process variables. In this context, the method according to the invention regulates the zinc layer thickness of a strip passing through a bath with liquid zinc by blowing air from a nozzle against the strip emerging from the zinc bath, the zinc layer thickness being the starting variable of the process, the air pressure being the manipulated variable and the belt speed and the geometric arrangement of the nozzle with respect to the belt are the influencing variables influencing the process.

Entsprechend den vorstehend angegebenen Ausbildungen des erfindungsgemäßen Verfahrens weist die Vorrichtung zur Führung der Beschichtungsanlage entsprechende, zur Durchführung des erfindungsgemäßen Verfahrens ausgebildete Mittel, d. h. insbesondere eine entsprechend programmierte Recheneinrichtung, auf.In accordance with the embodiments of the method according to the invention specified above, the device for guiding the coating system has corresponding means designed for carrying out the method according to the invention, i. H. in particular an appropriately programmed computing device.

Im folgenden wird die Erfindung unter Bezugnahme auf die Figuren näher erläutert. Dabei zeigen:

FIG 1
das Schema einer Feuerverzinkungsanlage,
FIG 2
das Blockschaltbild einer ersten Ausführungsform der erfindungsgemäßen Vorrichtung,
FIG 3
das Blockschaltbild einer zweiten Ausführungsform der erfindungsgemäßen Vorrichtung,
FIG 4
ein Blockschaltbild der um eine Regeleinrichtung erweiterten erfindungsgemäßen Vorrichtung,
FIG 5
ein Diagramm mit Kennlinien, die die Abhängigkeit der Ausgangsgröße des Prozesses von Einflußgrößen beschreiben,
FIG 6
ein Blockschaltbild eines Modells des Prozesses, bestehend aus einem Grundmodell und einer Korrektureinrichtung,
FIG 7
ein Beispiel für die Ausbildung des Grundmodells als neuronales Netzwerk,
FIG 8
ein Diagramm zur Erläuterung der Erzeugung einer Korrekturkomponente durch die Korrektureinrichtung und
FIG 9
ein Beispiel für die Ausbildung der Korrektureinrichtung als neuronales Netzwerk.
The invention is explained in more detail below with reference to the figures. Show:
FIG. 1
the scheme of a hot-dip galvanizing plant,
FIG 2
the block diagram of a first embodiment of the device according to the invention,
FIG 3
the block diagram of a second embodiment of the device according to the invention,
FIG 4
2 shows a block diagram of the device according to the invention expanded by a control device,
FIG 5
a diagram with characteristic curves that describe the dependence of the output variable of the process on influencing variables,
FIG 6
1 shows a block diagram of a model of the process, consisting of a basic model and a correction device,
FIG 7
an example for the formation of the basic model as a neural network,
FIG 8
a diagram for explaining the generation of a correction component by the correction device and
FIG. 9
an example of the formation of the correction device as a neural network.

FIG 1 zeigt das Schema einer Feuerverzinkungsanlage, in der ein gewalztes Band 1 mit einer Geschwindigkeit v ein Bad 2 mit flüssigem Zink durchläuft. Nach dem Austritt aus dem Zinkbad 2 bleibt eine noch flüssige Zinkschicht an dem Band 1 haften. Im Abstand a vom Band 1 befindet sich eine Düse 3, aus der mit einem Druck p Luft gegen das Band 1 strömt und so überschüssiges Zink entfernt. Der Luftdruck p wird dabei über ein Stellglied 4, hier ein Ventil, in Abhängigkeit von einer Stellgröße p* eingestellt. Im Verlauf des Bandes 1 ist hinter der Düse 3 ein Heißmeßgerät 5 angeordnet, mit dem die Zinkschichtdicke c auf dem Band 1 gemessen wird.1 shows the diagram of a hot-dip galvanizing plant in which a rolled strip 1 passes through a bath 2 with liquid zinc at a speed v. After leaving the zinc bath 2, a still liquid zinc layer adheres to the belt 1. At a distance a from the belt 1 there is a nozzle 3, from which air flows against the belt 1 at a pressure p and thus removes excess zinc. The air pressure p is set via an actuator 4, here a valve, as a function of an actuating variable p *. In the course of the strip 1, a hot measuring device 5 is arranged behind the nozzle 3, with which the zinc layer thickness c on the strip 1 is measured.

Haupteinflüsse für die Zinkschichtdicke c sind die Geschwindigkeit v des Bandes 1, der Luftdruck p, der Abstand a zwischen dem Band 1 und der Düse 3, die Höhe h der Düse 3 über dem Zinkbad 2 und der Winkel α zwischen dem Band 1 und der Düse 3.Main influences for the zinc layer thickness c are the speed v of the belt 1, the air pressure p, the distance a between the belt 1 and the nozzle 3, the height h of the nozzle 3 above the zinc bath 2 and the angle α between the belt 1 and the nozzle 3rd

Der in FIG 1 gezeigte technische Prozeß, bei dem sich die Ausgangsgröße des Prozesses, nämlich die Zinkschichtdicke c, in Abhängigkeit von der Stellgröße p* und den übrigen Einflußgrößen v, a, h und α einstellt, ist in dem Blockschaltbild nach FIG 2 in Form eines Funktionsblockes 6 dargestellt. Ziel ist es, die Stellgröße p* so einzustellen, daß die Ausgangsgröße c des Prozesses, also die tatsächliche gemessene Zinkschichtdicke, den Wert einer Führungsgröße c*, hier der Soll-Zinkschichtdicke, annimmt. Die Einstellung der Stellgröße p* erfolgt dabei im Unterschied zu einer herkömmlichen Regelung durch ein inverses Modell 7 des Prozesses 6, das als Eingangsgrößen die Führungsgröße c* und die Einflußgrößen v, a, h und α erhält und aus diesen Größen die Stellgröße p* direkt berechnet. Das inverse Modell 7 ist unmittelbar aus einem Modell 8 des Prozesses 6 abgeleitet, das den Prozeß 6 nachbildet und dementsprechend ebenso wie der Prozeß als Eingangsgrößen die Stellgröße p* und die Einflußgrößen v, a, h und α erhält. Diese Eingangsgrößen werden in dem Modell 8 über änderbare Modellparameter m zu einem Modellausgangswert cM verknüpft, der umso mehr der realen Ausgangsgröße c des Prozesses 6 entspricht, je genauer der Prozeß 6 durch das Modell 8 nachgebildet wird. Dazu wird das aufgrund von bereits vorhandenem Wissen über den Prozeß 6 voradaptierte Modell 8 nach Inbetriebnahme der gesamten Vorrichtung zur Führung des Prozesses 6 adaptiv an das tatsächliche Prozeßgeschehen angepaßt. Hierzu wird der Modellausgangswert cM in einer Vergleichseinrichtung 9 mit der gemessenen Ausgangsgröße c des Prozesses 6 verglichen und die Abweichung cM-c einem durch den Schaltungsblock 10 repräsentierten Lernalgorithmus zugeführt, der auf die Modellparameter m in dem Modell 8 zugreift und diese in Abhängigkeit von der ermittelten Abweichung cM-c im Sinne einer Verringerung dieser Abweichung verändert. Gleichzeitig werden auch in dem inversen Modell 7, das im Vergleich zu dem Modell 8 lediglich eine inverse Berechnung durchführt und deswegen direkt aus dem Modell 7 abgeleitet ist und auch dieselben Modellparameter m aufweist, die Modellparameter m verändert, so daß das Modell 8 und das zugehörige inverse Modell 7 beide an das tatsächliche Prozeßgeschehen angepaßt werden.The technical process shown in FIG. 1, in which the output variable of the process, namely the zinc layer thickness c, is set as a function of the manipulated variable p * and the other influencing variables v, a, h and α, is in the form of the block diagram according to FIG. 2 a functional block 6 shown. The aim is to set the manipulated variable p * so that the output variable c of the process, i.e. the actual measured zinc layer thickness, assumes the value of a reference variable c *, here the target zinc layer thickness. The setting In contrast to conventional regulation, the manipulated variable p * is carried out by an inverse model 7 of process 6, which receives the command variable c * and the influencing variables v, a, h and α as input variables and calculates the manipulated variable p * directly from these variables. The inverse model 7 is derived directly from a model 8 of the process 6, which simulates the process 6 and accordingly, like the process, receives the manipulated variable p * and the influencing variables v, a, h and α as input variables. In model 8, these input variables are linked via modifiable model parameters m to form a model output value c M , which corresponds more closely to the real output variable c of process 6, the more precisely model 6 replicates process 6. For this purpose, the model 8 pre-adapted on the basis of existing knowledge about the process 6 is adapted adaptively to the actual process after the entire device for guiding the process 6 has been put into operation. For this purpose, the model output value c M is compared in a comparison device 9 with the measured output variable c of the process 6 and the deviation c M -c is fed to a learning algorithm represented by the circuit block 10, which accesses the model parameters m in the model 8 and these as a function of the determined deviation c M -c changed in the sense of a reduction of this deviation. At the same time, the model parameters m are also changed in the inverse model 7, which only performs an inverse calculation in comparison to the model 8 and is therefore derived directly from the model 7 and also has the same model parameters m, so that the model 8 and the associated model inverse model 7 are both adapted to the actual process.

Bei guter Adaption des Modells 8 an den Prozeß 6 wird von dem inversen Modell 7 die Stellgröße p* in Abhängigkeit von der Führungsgröße c* und den Einflußgrößen v, a, h und α so genau berechnet, daß die Ausgangsgröße c des Prozesses den Wert der Führungsgröße c* annimmt. Dabei erfolgt die Berechnung der Stellgröße p* schneller, als dies bei einer herkömmlichen Regelung, die im übrigen auch ständig an den tatsächlichen Prozeßverlauf angepaßt werden muß, der Fall ist.With a good adaptation of the model 8 to the process 6, the manipulated variable p * is calculated by the inverse model 7 as a function of the reference variable c * and the influencing variables v, a, h and α so precisely that the output variable c of the process has the value of Takes c *. The manipulated variable p * is calculated faster than with a conventional control, which also has to be constantly adapted to the actual process, is the case.

FIG 3 zeigt ein alternatives Ausführungsbeispiel für die erfindungsgemäße Vorrichtung zur Führung des Prozesses 6. Dieses Ausführungsbeispiel unterscheidet sich von dem in FIG 2 durch das Fehlen des inversen Modells 7. Stattdessen wird der Modellausgangswert cM in einer weiteren Vergleichseinrichtung 11 mit der Führungsgröße c* verglichen und das Vergleichsergebnis c*-cM einer Einrichtung 12 zugeführt, die ausgangsseitig einen Wert für die Stellgröße p* erzeugt und diesen Wert so lange verändert, wie das Vergleichsergebnis c*-cM nicht Null ist. Da der von der Einrichtung 12 erzeugte und veränderte Wert für die Stellgröße p* dem Modell 8 zugeführt wird, ändert sich dessen Ausgangswert cM so lange, bis er denselben Wert wie die Führungsgröße c* aufweist. Wenn dies der Fall ist, wird in Abhängigkeit von dem Vergleichsergebnis c*-cM = 0 der aktuelle Wert für die Stellgröße p* über eine Schalteinrichtung 13 an das Stellglied 4 (FIG 1) durchgeschaltet.3 shows an alternative exemplary embodiment for the device according to the invention for guiding process 6. This exemplary embodiment differs from that in FIG. 2 by the lack of the inverse model 7. Instead, the model output value c M is compared in a further comparison device 11 with the reference variable c * and the comparison result c * -c M is fed to a device 12 which generates a value for the manipulated variable p * on the output side and changes this value as long as the comparison result c * -c M is not zero. Since the value for the manipulated variable p * generated and changed by the device 12 is fed to the model 8, its output value c M changes until it has the same value as the command variable c *. If this is the case, depending on the comparison result c * -c M = 0, the current value for the manipulated variable p * is switched through to the actuator 4 (FIG. 1) via a switching device 13.

Wie bereits erwähnt, wird mit den in FIG 2 und 3 gezeigeten Ausführungsbeispielen der erfindungsgemäßen Vorrichtung bei hinreichend genauer Adaption des Modells 8 an das Prozeßgeschehen eine exakte Führung des Prozesses 6 erreicht, ohne daß eine Regeleinrichtung erforderlich ist. Wenn die Nachbildung des Prozesses 6 durch das Modell 8 als nicht hinreichend genau angesehen wird, kann, wie FIG 4 zeigt, zusätzlich eine Regeleinrichtung 14 vorgesehen werden. In FIG 4 ist mit 15 eine Recheneinrichtung bezeichnet, in der die in den FIG 2 bzw. 3 gezeigten Schaltungsblöcke 7 bis 13 implementiert sind. Der Regeleinrichtung 14 wird eingangsseitig die in einer zusätzlichen Vergleichseinrichtung 16 gebildete Regelabweichung Δc zwischen der Führungsgröße c* oder, wie durch die gestrichelte Linie angedeutet ist, dem Modellausgangswert cM und der Ausgangsgröße c des Prozesses zugeführt. Das Ausgangssignal Δp der Regeleinrichtung 14 wird über ein Additionsglied 17 der Stellgröße p* aufgeschaltet. Die Recheneinrichtung 15 mit dem Modell 8 bildet also eine Vorsteuerung für die Regeleinrichtung 14. Die Regeleinrichtung 14 muß daher nur noch solche Regelfehler ausregeln, die von dem Modell 8 nicht erfaßt werden, so daß die Regeleinrichtung 14 nur in geringem Maße beansprucht wird und dementsprechend einfach ausgebildet sein kann. Um zu erreichen, daß die Regeleinrichtung 14 angepaßt an das reale Prozeßgeschehen immer im optimalen Arbeitspunkt arbeitet, wird die Verstärkung ∂c/∂p der Regeleinrichtung 14 mittels der Änderungen der von dem Modell 8 verarbeiteten Größen cM und p* eingestellt.As already mentioned, with the exemplary embodiments of the device according to the invention shown in FIGS. 2 and 3, with a sufficiently precise adaptation of the model 8 to the process, an exact control of the process 6 is achieved without a control device being required. If the simulation of process 6 by model 8 is not considered to be sufficiently precise, a control device 14 can additionally be provided, as shown in FIG. In FIG. 4, 15 denotes a computing device in which the circuit blocks 7 to 13 shown in FIGS. 2 and 3 are implemented. The control deviation 14 is supplied on the input side with the control deviation Δc formed in an additional comparison device 16 between the command variable c * or, as indicated by the broken line, the model output value c M and the output variable c of the process. The output signal Δp of the control device 14 is applied to the manipulated variable p * via an adder 17. The computing device 15 with the model 8 thus forms a precontrol for the control device 14. The control device 14 therefore only has to correct those control errors which are not covered by the model 8, so that the control device 14 is only stressed to a small extent and is accordingly simple can be trained. In order to ensure that the control device 14 always works at the optimum working point, adapted to the real process, the gain ∂c / ∂p of the control device 14 is set by means of the changes in the quantities c M and p * processed by the model 8.

Im folgenden wird der Aufbau des Modells 8 und der Lernvorgang zur Anpassung des Modells 8 an das Prozeßgeschehen näher erläutert. Um die Darstellung zu vereinfachen, wird nur der Einfluß der Bandgeschwindigkeit v und des Luftdruckes p auf die Zinkschichtdicke c betrachtet.The structure of the model 8 and the learning process for adapting the model 8 to the process are explained in more detail below. To simplify the illustration, only the influence of the belt speed v and the air pressure p on the zinc layer thickness c is considered.

FIG 5 zeigt die nichtlinearen Kennlinien 18 für die Zinkschichtdicke c in Abhängigkeit von der Bandgeschwindigkeit v und dem Luftdruck p. Diese Kennlinien sollen durch das Modell 8 des Prozesses nachgebildet werden. Wie FIG 6 zeigt, besteht das Modell 8 aus einem Grundmodell 19 und einer dazu parallelen Korrektureinrichtung 20, denen jeweils die Eingangsgrößen p und v zugeführt werden. Das Grundmodell 19 berechnet daraus eine Grundkomponente cMF und die Korrektureinrichtung 20 eine Korrekturkomponente cMK, die beide in einem Additionsglied 21 zu dem Modellausgangswert cM verknüpft werden.5 shows the non-linear characteristic curves 18 for the zinc layer thickness c as a function of the strip speed v and the air pressure p. These characteristics should be modeled by model 8 of the process. As FIG 6 shows, the model 8 consists of a basic model 19 and a correction device 20 parallel thereto, to which the input variables p and v are fed in each case. The basic model 19 uses this to calculate a basic component c MF and the correction device 20 a correction component c MK , both of which are linked in an adder 21 to the model output value c M.

Bei der Grundkomponente cMF handelt es sich um eine Funktionsapproximation auf der Grundlage von vorhandenem Wissen über den Prozeß. Dieses Wissen besteht aus Meßdaten, die den Zusammenhang zwischen der Zinkschichtdicke c und den auf sie wirkenden Einflußgrößen p, v, a, h und α beschreiben und durch Messungen an einer ausgewählten Feuerverzinkungsanlage erhalten werden. Aus der graphischen Darstellung der Meßdaten, die dem Verlauf der Kennlinien 18 in FIG 5 entspricht, wird eine mathematische Funktion cMF = f (p, v, h, α) abgeleitet, die den prinzipiellen Verlauf der Kennlinien 18 annähernd wiedergibt. Sind a, h und α konstant, so lassen sich die Kennlinien 18 beispielsweise durch die Funktion c MF = f(p, v) = v/(K1·v+K2·p 3/2 +K3)

Figure imgb0001
approximieren, wobei K1, K2 und K3 zu lernende Funktionsparameter sind.The basic component c MF is a function approximation based on existing knowledge of the process. This knowledge consists of measurement data that describe the relationship between the zinc layer thickness c and the influencing variables p, v, a, h and α that affect it and are obtained by measurements on a selected hot-dip galvanizing plant. From the graphical representation of the measurement data, which corresponds to the course of the characteristic curves 18 in FIG. 5, a mathematical function c MF = f (p, v, h, α) is derived, which approximately reproduces the basic course of the characteristic curves 18. If a, h and α are constant, the characteristic curves 18 can be determined, for example, by the function c MF = f (p, v) = v / (K1v + K2p 3/2 + K3)
Figure imgb0001
approximate, whereby K1, K2 and K3 are function parameters to be learned.

Die so erhaltene mathematische Funktion cMF = f(p, v) wird in Form eines neuronalen Netzwerkes 22 realisiert, das in FIG 7 dargestellt ist. Dem neuronalen Netzwerk 22 werden eingangsseitig die Größen v und p3/2 zugeführt und entsprechend der oben angegebenen mathematischen Funktion durch einen Summationsknoten 23 und einen Knoten 24 mit Quotientenfunktion verknüpft. Die Verbindungen der Knoten in dem neuronalen Netzwerk 22 sind jeweils mit einem Faktor versehen. Ist die Verbindung dick gezeichnet, so ist der Faktor lernbar (Netzwerkparameter K1, K2, K3); ansonsten ist der Faktor konstant (hier 1,0). An dem Ausgang des neuronalen Netzwerkes 22 ergibt sich die Grundkomponente cMF.The mathematical function c MF = f (p, v) obtained in this way is implemented in the form of a neural network 22, which is shown in FIG. The variables v and p 3/2 are fed to the neural network 22 on the input side and are linked according to the mathematical function specified above by a summation node 23 and a node 24 with quotient function. The connections of the nodes in the neural network 22 are each provided with a factor. If the connection is drawn thick, the factor can be learned (network parameters K1, K2, K3); otherwise the factor is constant (here 1.0). The basic component c MF results at the output of the neural network 22.

Um die Netzwerkparameter K1, K2 und K3 zu lernen, wird die erzeugte Grundkomponente cMF mit der von dem Prozeß 6 in Abhängigkeit von den gemessenen Größen p und v erzeugten und gemessenen Ausgangsgröße c in einer Vergleichseinrichtung 25 verglichen. In Abhängigkeit von dem Vergleichsergebnis werden in einem Lernalgorithmus 26 die Netzwerkparameter K1, K2 und K3 solange verändert, bis die Angleichung der Grundkomponente cMF an die Ausgangsgröße c des Prozesses 6 nicht weiter verbessert werden kann.In order to learn the network parameters K1, K2 and K3, the generated basic component c MF is compared in a comparison device 25 with the output variable c generated and measured by the process 6 as a function of the measured variables p and v. Depending on the comparison result, the network parameters K1, K2 and K3 are changed in a learning algorithm 26 until the approximation of the basic component c MF to the output variable c of the process 6 cannot be further improved.

Die von dem so adaptierten Grundmodell 19 erzeugte Grundkomponente cMF ist in FIG 5 gestrichelt dargestellt. Wie zu sehen ist, ergeben sich immer noch Differenzen zwischen der Ausgangsgröße c des Prozesses 6 und der von dem Grundmodell 19 gelieferten Grundkomponente cMF. Zum Ausgleich dieser Differenzen dient die Korrekturkomponente cMK, deren Berechnung in der Korrektureinrichtung 20 anhand von FIG 8 näher erläutert wird. Zunächst werden in dem von den auf den Prozeß einwirkenden Größen p, v, a, h und α definierten mehrdimensionalen Raum Stützstellen definiert. Zur Vereinfachung werden hier, wie auch schon vorher, nur die Größen p und v betrachtet, wobei die Stützstellen durch die Wertepaare (pi, vj) mit i, j = 0,1,2,... definiert werden. Für diese Stützstellen (pi, vj) werden nun Korrekturwerte cMKij gelernt, indem die Differenz ΔcM zwischen der in Abhängigkeit von den gemessenen Größen p und v von dem Prozeß 6 erzeugten Ausgangsgröße c und der von dem Grundmodell 19 gelieferten Grundkomponente cMF zur Berechnung der stützstellenbezogenen Korrekturwerte cMKij herangezogen wird. Mit anderen Worten werden die stützstellenbezogenen Korrekturwerte cMKij derart aus dem Differenzwert ΔcM berechnet, daß sich umgekehrt bei Interpolation der Korrekturwerte cMKij an den Stützstellen (pi, vj) der Differenzwert ΔcM für die aktuellen Eingangsgrößen p und v ergibt.The basic component c MF generated by the basic model 19 thus adapted is shown in dashed lines in FIG. As can be seen, there are still differences between the output variable c of the process 6 and the basic component c MF provided by the basic model 19. The correction component c MK serves to compensate for these differences, and its calculation is explained in more detail in the correction device 20 on the basis of FIG. First, support points are defined in the multidimensional space defined by the variables p, v, a, h and α that act on the process. For the sake of simplicity, as before, only the quantities p and v are considered, with the reference points being defined by the value pairs (p i , v j ) with i, j = 0,1,2, .... Correction values c MKij are now learned for these reference points (p i , v j ) by the difference Δc M between the output variable c generated by the process 6 as a function of the measured variables p and v and the basic component c MF supplied by the basic model 19 is used to calculate the base point-related correction values c MKij . In other words, the reference point-related correction values c MKij are calculated from the difference value Δc M in such a way that, when the correction values c MKij are interpolated at the reference points (p i , v j ), the difference value Δc M results for the current input variables p and v.

Durch Interpolation zwischen den so gelernten stützstellenbezogenen Korrekturwerten cMKij wird schließlich die Korrekturkomponente cMK berechnet. Die Interpolation erfolgt auf der Grundlage von Fuzzy-Folgerungen, wozu jeder Stützstelle (pi, vj) jeweils eine Zuordnungs- oder Membership-Funktion 27 zugeordnet ist, die den gelernten Korrekturwert cMKij an der betreffenden Stützstelle und in deren Umgebung mit einem Gewicht w wichtet. Dabei haben die Zuordnungsfunktionen 27 an den zugehörigen Stützstellen (pi, vj) den Wert 1, der bis zu den jeweils benachbarten Stützstellen linear auf den Wert 0 abfällt. Die Korrekturkomponente cMK für die Eingangsgrößen p und v berechnet sich aus der Addition aller zu den Eingangsgrößen p und v benachbarten und mit dem jeweiligen Gewicht w multiplizierten stützstellenbezogenen Korrekturwerte cMKij mit c MK = w 1 w 2 c MK00 +w 1 w 4 c MK10 +w 2 w 3 c MK01 +w 3 w 4 c MK11 .

Figure imgb0002
The correction component c MK is finally calculated by interpolation between the correction point values c MKij learned in this way. The interpolation takes place on the basis of fuzzy inferences, for which each reference point (p i , v j ) is assigned an assignment or membership function 27, which weights the learned correction value c MKij at the relevant reference point and in its vicinity w weights. The assignment functions 27 have the value 1 at the associated reference points (p i , v j ), which drops linearly to the value 0 up to the respectively adjacent reference points. The correction component c MK for the input variables p and v is calculated from the addition of all the reference point-related correction values c MKij that are adjacent to the input variables p and v and multiplied by the respective weight w c MK = w 1 w 2nd c MK00 + w 1 w 4th c MK10 + w 2nd w 3rd c MK01 + w 3rd w 4th c MK11 .
Figure imgb0002

FIG 9 zeigt ein Beispiel für die Realisierung der oben stehenden Berechnungsvorschrift für die Korrekturkomponente cMK durch ein neuronales Netzwerk 28. Dieses erhält als Eingangsgrößen die Größen v und p, die einer ersten Schicht mit Knoten 29 mit Sigmoid-Funktionen zugeführt werden. Dabei werden die in FIG 8 gezeigten dreieckförmigen Zugehörigkeits-Funktionen 27 durch jeweils zwei der Knoten 29 und nachfolgende Addition der Knotenausgangssignale in jeweils einem Summationsknoten 30 gebildet. An den Ausgängen der Summationsknoten 30 erscheinen daher die Werte w1 bis w4, die die mit den Zugehörigkeitsfunktionen 27 gewichteten Größen p und v bezeichnen. Diese Werte w1 bis w4 werden entsprechend der obigen Rechenvorschrift durch Knoten 31 mit Produktfunktion und einen Summationsknoten 32 am Ausgang des neuronalen Netzwerkes 28 zu der Korrekturkomponente cMK verknüpft. Die Faktoren cMK00 bis cMK11 an den Eingängen des Summationsknotens 32 sind die zu lernenden Netzwerkparameter des neuronalen Netzwerkes 28.9 shows an example of the implementation of the above calculation rule for the correction component c MK through a neural network 28. This receives as inputs the variables v and p, which are fed to a first layer with nodes 29 with sigmoid functions. The triangular membership functions 27 shown in FIG. 8 are each formed by two of the nodes 29 and subsequent addition of the node output signals in a summation node 30 each. The values w 1 to w 4 , which denote the quantities p and v weighted with the membership functions 27, therefore appear at the outputs of the summation nodes 30. These values w 1 to w 4 are linked according to the above calculation rule by node 31 with product function and a summation node 32 at the output of the neural network 28 to the correction component c MK . The factors c MK00 to c MK11 at the inputs of the summation node 32 are the network parameters of the neural network 28 to be learned.

Claims (12)

  1. Method to control a coating system to coat a carrier material, for example steel plates, with a coating material, for example zinc, whereby coating material is applied to the carrier and is subsequently removed according to a specified desired layer thickness (c*) by means of air under pressure (p) which flows against the carrier, whereby by means of a model of the coating system (6) contained in a computing device (15), in dependence upon a desired air pressure (p*) supplied to the model (8) on the input side and upon influencing variables (v, a, h, α), on which the actual layer thickness (c) depends, and in dependence upon variable model parameters (m), a model output value (cM) is calculated, whereby the model output value (cM) is compared with the measured actual layer thickness and, in dependence upon the comparison result, the model parameters (m) are changed in the sense of a reduction of the difference between the model output value (cM) and the actual layer thickness (c), and whereby the model (8) of the coating system adjusted in this way to the actual process events is used to determine the desired air pressure (p*) in dependence upon the influencing variables (v, a, h, α), characterized in that the model (8) of the coating system adjusted to the actual process events is used to determine the desired air pressure (p*) in dependence upon the desired layer thickness (c*), in that to determine the desired air pressure (p*) the desired layer thickness (c*) is compared with the model output value (cM) and, in dependence upon the comparison result, a value supplied to the model (8) for the desired air pressure (p*) is changed until the model output value (cM) corresponds to the desired layer thickness (c*) and in that the value finally obtained for the desired air pressure (p*) is supplied to an actuator (4).
  2. Method to control a coating system to coat a carrier material, for example steel plates, with a coating material, for example zinc, whereby coating material is applied to the carrier and is subsequently removed according to a specified desired layer thickness (c*) by means of air under pressure (p) which flows against the carrier, whereby by means of a model of the coating system (6) contained in a computing device (15), in dependence upon a desired air pressure (p*) supplied to the model (8) on the input side and upon influencing variables (v, a, h, α), on which the actual layer thickness (c) depends, and in dependence upon variable model parameters (m), a model output value (cM) is calculated, whereby the model output value (cM) is compared with the measured actual layer thickness and, in dependence upon the comparison result, the model parameters (m) are changed in the sense of a reduction of the difference between the model output value (cM) and the actual layer thickness (c), and whereby the model (8) of the coating system adjusted in this way to the actual process events is used to determine the desired air pressure (p*) in dependence upon the influencing variables (v, a, h, α), characterized in that the model (8) of the coating system adjusted to the actual process events is used to determine the desired air pressure (p*) in dependence upon the desired layer thickness (c*), in that to determine the desired air pressure (p*) the desired layer thickness (c*) and the influencing variables (v, a, h, α) are supplied on the input side to an inverse model (7) of the coating system (6), which generates the desired air pressure (p*) on the output side, and in that in dependence upon the adjustment of the model (8) of the coating system (6) the inverse model (7) is also adjusted to the process events.
  3. Method according to one of the preceding claims, characterized in that the system deviation (Δc) between the model output value (cM) or the specified desired layer thickness (c*) and the measured actual layer thickness (c) is supplied to a regulating device (14), the output signal (Δp) of which is applied to the desired air pressure (p*) generated by means of the model (8) of the coating system (6).
  4. Method according to claim 3, characterized in that the amplification of the regulating device (14) is adjusted by means of the changes of the variables (cM, p*) processed by the model (8), which changes are supplied by the model (8) of the coating system (6).
  5. Method according to one of the preceding claims, characterized in that the model output value (cM) is formed by linking a base component (cMF) and a correction component (cMK), in that the base component (cMF) is generated by a basic model (19) of the coating system (6), which basic model has been created on the basis of existing knowledge of the coating system (6) and is adjusted to the process events with the aid of the measured actual layer thickness (c), and in that the correction component (cMK) is generated by a correction device (20) which, in dependence upon differences between the model output value (cM) and the actual layer thickness (c) which still exist after the adjustment of the basic model (19) has taken place, is adaptively adjusted to the process events.
  6. Method according to claim 5, characterized in that to create the basic model (19) of the coating system (6), on the basis of existing knowledge of the coating system (6) a mathematical function is created which reproduces the dependence of the actual layer thickness (c) on the desired air pressure (p*), the influencing variables (v, a, h, α) and on function parameters (K1, K2, K3), and in that the function obtained in this way is realized as a neuronal network (20), to which on the input side the desired air pressure (p*) and the influencing variables (v, a, h, α) are supplied, and the network parameters (K1, K2, K3) of which, corresponding to the function parameters, are adaptively adjusted to the process events.
  7. Method according to claim 5 or 6, characterized in that the correction device (20) for specified supporting points (pi, vj) prepares supporting-point-related correction values (cMKij) in the multi-dimensional area defined by the desired air pressure (p*) and the influencing variables (v, a, h, α), in that the supporting-point-related correction values (cMKij) are calculated from differences between the model output value (cM) and the actual layer thickness (c) of the coating system (6) measured together with values of the desired air pressure (p*) and the influencing variables (v, a, h, α) lying between the supporting points (pi, vj), in that upon interpolation of the supporting-point-related correction values (cMKij) the respectively measured difference (ΔcM) again results, and in that the correction component (cMK) is calculated by means of interpolation between those supporting-point-related correction values (cMKij), the supporting points (pi, vj) of which are adjacent to the values for the desired air pressure (p*) belonging to the correction component (cMK) and the influencing variables (v, a, h, α).
  8. Method according to claim 7, characterized in that the interpolation to calculate the correction component (cMK) takes place on the basis of fuzzy conclusions, in that each supporting point (pi, vj) is allocated an allocation (membership) function (27) which at the relevant supporting point (pi, vj) has the value one and falls up to the next adjacent supporting points to the value zero, and in that the supporting-point-related correction values (cMKij) are weighted with the associated allocation functions (27) and are superimposed, one on the other.
  9. Method according to one of claims 5 to 8, characterized in that the correction component (cMK) is calculated in a neuronal network (28) which forms the correction device (20).
  10. Method according to one of the preceding claims, characterized in that in the coating system (6) the actual layer thickness (c) of zinc on a strip (1) which passes through a bath (2) of liquid zinc is regulated, in that air is blown from a nozzle (3) against the strip (1) when it leaves the zinc bath (2), whereby the air pressure (p), the strip speed (v) and the geometric arrangement of the nozzle (3) with respect to the strip (1) are the influencing variables which act on the coating system (6).
  11. Device to control a coating system to coat a carrier material, for example steel plates, with a coating material, for example zinc, whereby coating material is applied to the carrier, and is subsequently removed according to a specified desired layer thickness (c*) by means of air under pressure (p) which flows against the carrier, whereby the device has a computing device (15) on which a model of the coating system (6) is implemented and which, in dependence upon a desired air pressure (p*) supplied to the model (8) on the input side and upon influencing variables (v, a, h, α), on which the actual layer thickness (c) depends, and in dependence upon variable model parameters (m), is constructed to calculate a model output value (cM), whereby it is constructed to compare the model output value (cM) with the measured actual layer thickness and, in dependence upon the comparison result, to change the model parameters (m) in the sense of a reduction of the difference between the model output value (cM) and the actual layer thickness (c) and whereby the model (8) of the coating system adjusted in this way to the actual process events is used to determine the desired air pressure (p*) in dependence upon the influencing variables (v, a, h, α), characterized in that the device to control the coating system uses the model (8) of the coating system adjusted to the actual process events to determine the desired air pressure (p*) in dependence upon the desired layer thickness (c*), in that to determine the desired air pressure (p*) it compares the desired layer thickness (c*) with the model output value (cM) and, in dependence upon the comparison result, changes a value for the desired air pressure (p*) supplied to the model (8) until the model output value (cM) corresponds to the desired layer thickness (c*) and in that it supplies the finally obtained value for the desired air pressure (p*) to an actuator (4).
  12. Device to control a coating system to coat a carrier material, for example steel plates, with a coating material, for example zinc, whereby coating material is applied to the carrier and is subsequently removed according to a specified desired layer thickness (c*) by means of air under pressure (p) which flows against the carrier, whereby the device has a computing device (15) on which a model of the coating system (6) is implemented and which, in dependence upon a desired air pressure (p*) supplied to the model (8) on the input side and upon influencing variables (v, a, h, α), on which the actual layer thickness (c) depends, and in dependence upon variable model parameters (m), is constructed to calculate a model output value (cM), whereby it is constructed to compare the model output value (cM) with the measured actual layer thickness 5 and, in dependence upon the comparison result, to change the model parameters (m) in the sense of a reduction of the difference between the model output value (cM) and the actual layer thickness (c), and whereby the model (8) of the coating system adjusted in this way to the actual process events is used to determine the desired air pressure (p*) in dependence upon the influencing variables (v, a, h, α), characterized in that the device to control the coating system uses the model (8) of the coating system adjusted to the actual process events to determine the desired air pressure (p*) in dependence upon the desired layer thickness (c*), and in that to determine the desired air pressure (p*) it supplies the desired layer thickness (c*) and the influencing variables (v, a, h, α) on the input side to an inverse model (7) of the coating system (6), which generates the desired air pressure (p*) on the output side, and in that in dependence upon the adjustment of the model (8) of the coating system (6) the inverse model (7) is also adjusted to the process events.
EP94100589A 1994-01-17 1994-01-17 Method and apparatus for controlling a process Expired - Lifetime EP0663632B1 (en)

Priority Applications (5)

Application Number Priority Date Filing Date Title
EP94100589A EP0663632B1 (en) 1994-01-17 1994-01-17 Method and apparatus for controlling a process
DE59404777T DE59404777D1 (en) 1994-01-17 1994-01-17 Process and device for carrying out a process
AT94100589T ATE161109T1 (en) 1994-01-17 1994-01-17 METHOD AND DEVICE FOR CONDUCTING A PROCESS
US08/373,819 US5598329A (en) 1994-01-17 1995-01-17 Method and device for controlling a process
KR1019950000666A KR100354410B1 (en) 1994-01-17 1995-01-17 Process control method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
EP94100589A EP0663632B1 (en) 1994-01-17 1994-01-17 Method and apparatus for controlling a process

Publications (2)

Publication Number Publication Date
EP0663632A1 EP0663632A1 (en) 1995-07-19
EP0663632B1 true EP0663632B1 (en) 1997-12-10

Family

ID=8215610

Family Applications (1)

Application Number Title Priority Date Filing Date
EP94100589A Expired - Lifetime EP0663632B1 (en) 1994-01-17 1994-01-17 Method and apparatus for controlling a process

Country Status (5)

Country Link
US (1) US5598329A (en)
EP (1) EP0663632B1 (en)
KR (1) KR100354410B1 (en)
AT (1) ATE161109T1 (en)
DE (1) DE59404777D1 (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO1999031949A2 (en) * 1997-12-19 1999-07-01 Siemens Aktiengesellschaft Method and device for coating a metal strip
FR2913432A1 (en) * 2007-03-07 2008-09-12 Siemens Vai Metals Tech Sas METHOD AND INSTALLATION FOR CONTINUOUS DEPOSITION OF A COATING ON A TAPE SUPPORT

Families Citing this family (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5913184A (en) * 1994-07-13 1999-06-15 Siemens Aktiengesellschaft Method and device for diagnosing and predicting the operational performance of a turbine plant
DE19516627A1 (en) * 1995-05-05 1996-11-07 Ranco Inc Method and device for controlling a process
DE19618712B4 (en) * 1996-05-09 2005-07-07 Siemens Ag Control method for a roll stand for rolling a strip
DE29610789U1 (en) * 1996-06-19 1997-07-17 Siemens Ag Device for identifying a transmission path, in particular a control path
DE19642918C2 (en) * 1996-10-17 2003-04-24 Siemens Ag System for calculating the final thickness profile of a rolled strip
DE19745132A1 (en) * 1997-10-13 1999-04-15 Siemens Ag Method for coating a metal strip
JP2002501245A (en) * 1998-01-22 2002-01-15 エムティエス・システムズ・コーポレーション Method and apparatus for generating an input signal in a physical system
US7031949B2 (en) * 1998-01-22 2006-04-18 Mts Systems Corporation Method and apparatus for generating input signals in a physical system
US6285972B1 (en) 1998-10-21 2001-09-04 Mts Systems Corporation Generating a nonlinear model and generating drive signals for simulation testing using the same
DE69941339D1 (en) 1998-11-13 2009-10-08 Mts System Corp MEASURING THE REPEATABLE BANDWIDTH OF A SYSTEM FOR SIMULATION TESTING
US6317654B1 (en) 1999-01-29 2001-11-13 James William Gleeson Control of crude refining by a method to predict lubricant base stock's ultimate lubricant preformance
US6295485B1 (en) 1999-01-29 2001-09-25 Mobil Oil Corporation Control of lubricant production by a method to predict a base stock's ultimate lubricant performance
AU756690B2 (en) * 1999-01-29 2003-01-23 Mobil Oil Corporation Method to control a lubricant production
US6442445B1 (en) 1999-03-19 2002-08-27 International Business Machines Corporation, User configurable multivariate time series reduction tool control method
DE19939973A1 (en) * 1999-08-24 2001-03-01 Volkswagen Ag Regulation of a gasoline engine
WO2002025240A1 (en) 2000-09-21 2002-03-28 Mts Systems Corporation Multiple region convolver with tapering
DE10146791A1 (en) * 2001-09-20 2003-04-10 Sms Demag Ag Method and device for coating the surface of strand-like metallic material
JP4038501B2 (en) * 2003-09-02 2008-01-30 株式会社東芝 Inverse model calculation apparatus and inverse model calculation method
DE102004011236A1 (en) * 2004-03-04 2005-09-29 Bayerische Motoren Werke Ag Process control system
KR101502198B1 (en) * 2008-02-08 2015-03-12 지멘스 바이 메탈스 테크놀로지 에스에이에스 Method for the hardened galvanisation of a steel strip
KR102153149B1 (en) * 2012-10-24 2020-09-07 도쿄엘렉트론가부시키가이샤 Correction value computation device, correction value computation method, and computer program
US10247043B2 (en) * 2014-12-31 2019-04-02 General Electric Company Ducted cowl support for a gas turbine engine

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS60165399A (en) * 1984-02-07 1985-08-28 Hitachi Zosen Corp Plating thickness controlling method
JP2804318B2 (en) * 1989-11-29 1998-09-24 川崎製鉄株式会社 Plating weight control method
JP3221497B2 (en) * 1991-06-20 2001-10-22 三菱電機株式会社 Control method and power system voltage-reactive power control device using the control method
FR2678645B1 (en) * 1991-07-01 1993-10-29 Sollac METHOD FOR REGULATING A METALLURGICAL TREATMENT PERFORMED ON A RUNNING PRODUCT AND DEVICE FOR IMPLEMENTING SAME.
JP2804400B2 (en) * 1991-12-18 1998-09-24 川崎製鉄株式会社 Method for controlling coating weight of continuous hot-dip plating
US5368715A (en) * 1993-02-23 1994-11-29 Enthone-Omi, Inc. Method and system for controlling plating bath parameters
JPH06259107A (en) * 1993-03-02 1994-09-16 Kobe Steel Ltd Learning control method for process line

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO1999031949A2 (en) * 1997-12-19 1999-07-01 Siemens Aktiengesellschaft Method and device for coating a metal strip
WO1999031949A3 (en) * 1997-12-19 1999-08-26 Siemens Ag Method and device for coating a metal strip
FR2913432A1 (en) * 2007-03-07 2008-09-12 Siemens Vai Metals Tech Sas METHOD AND INSTALLATION FOR CONTINUOUS DEPOSITION OF A COATING ON A TAPE SUPPORT
WO2008110673A1 (en) * 2007-03-07 2008-09-18 Siemens Vai Metals Technologies Sas Method and equipment for the continuous deposition of a coating on a strip type substrate

Also Published As

Publication number Publication date
KR950033889A (en) 1995-12-26
EP0663632A1 (en) 1995-07-19
DE59404777D1 (en) 1998-01-22
ATE161109T1 (en) 1997-12-15
KR100354410B1 (en) 2002-12-16
US5598329A (en) 1997-01-28

Similar Documents

Publication Publication Date Title
EP0663632B1 (en) Method and apparatus for controlling a process
DE4416364B4 (en) Method and control device for regulating a process
DE4338615B4 (en) Method and device for managing a process in a controlled system
DE4338608B4 (en) Method and device for managing a process in a controlled system
DE4008510C2 (en) Control device with optimal decision unit
DE19522494C2 (en) Process for rolling a metal strip
EP0534221B1 (en) Control parameter improvement method for industrial installations
DE4338607B4 (en) Method and device for managing a process in a controlled system
DE4416317B4 (en) Method and control device for controlling a material processing process
DE102019214548A1 (en) CONTROL DEVICE AND CONTROL METHOD
EP0756219A2 (en) Method for supervising product properties and method for controlling a manufacturing process
WO1998015882A1 (en) A process and device for identification or pre-calculation of parameters of a time-variant industrial process
DE19881041B4 (en) Method for controlling and presetting a steelworks or parts of a steelworks
DE3832789A1 (en) TROUBLESHOOTING DEVICE
EP0815498A1 (en) Process controlling method and device
DE4239270A1 (en) Process for actuator identification in the cross-profile control of a continuously generated material web
WO1995019591A1 (en) Method and device for running a process
EP3542229A1 (en) System and method for determining the parameters of a controller
DE4319926C2 (en) Process for controlling a continuous process with an optimization and a control phase
EP0700536A1 (en) Regulating device
DE19748718A1 (en) Self-adjusting device for process control system
DE4333804C2 (en) Control device
DE19527521C1 (en) Neural network training method
DE4232752C1 (en) Process for generating a sharp output manipulated variable at the output of a fuzzy control loop
DE1538596C (en) Procedure and arrangement for self-adapting regulation

Legal Events

Date Code Title Description
PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

17P Request for examination filed

Effective date: 19950103

AK Designated contracting states

Kind code of ref document: A1

Designated state(s): AT BE DE FR GB IT

17Q First examination report despatched

Effective date: 19950703

GRAG Despatch of communication of intention to grant

Free format text: ORIGINAL CODE: EPIDOS AGRA

GRAG Despatch of communication of intention to grant

Free format text: ORIGINAL CODE: EPIDOS AGRA

GRAH Despatch of communication of intention to grant a patent

Free format text: ORIGINAL CODE: EPIDOS IGRA

GRAH Despatch of communication of intention to grant a patent

Free format text: ORIGINAL CODE: EPIDOS IGRA

GRAA (expected) grant

Free format text: ORIGINAL CODE: 0009210

AK Designated contracting states

Kind code of ref document: B1

Designated state(s): AT BE DE FR GB IT

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: IT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRE;WARNING: LAPSES OF ITALIAN PATENTS WITH EFFECTIVE DATE BEFORE 2007 MAY HAVE OCCURRED AT ANY TIME BEFORE 2007. THE CORRECT EFFECTIVE DATE MAY BE DIFFERENT FROM THE ONE RECORDED.SCRIBED TIME-LIMIT

Effective date: 19971210

REF Corresponds to:

Ref document number: 161109

Country of ref document: AT

Date of ref document: 19971215

Kind code of ref document: T

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: AT

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 19980117

REF Corresponds to:

Ref document number: 59404777

Country of ref document: DE

Date of ref document: 19980122

ET Fr: translation filed
GBT Gb: translation of ep patent filed (gb section 77(6)(a)/1977)

Effective date: 19980206

PLBE No opposition filed within time limit

Free format text: ORIGINAL CODE: 0009261

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: NO OPPOSITION FILED WITHIN TIME LIMIT

26N No opposition filed
REG Reference to a national code

Ref country code: GB

Ref legal event code: IF02

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: GB

Payment date: 20090116

Year of fee payment: 16

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: BE

Payment date: 20090122

Year of fee payment: 16

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: DE

Payment date: 20090320

Year of fee payment: 16

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: FR

Payment date: 20090120

Year of fee payment: 16

BERE Be: lapsed

Owner name: *SIEMENS A.G.

Effective date: 20100131

GBPC Gb: european patent ceased through non-payment of renewal fee

Effective date: 20100117

REG Reference to a national code

Ref country code: FR

Ref legal event code: ST

Effective date: 20100930

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: FR

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20100201

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: DE

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20100803

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: GB

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20100117

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: BE

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20100131